Struct VectorIndexCoordinator 

pub struct VectorIndexCoordinator<E: StorageEngine> { /* private fields */ }

Coordinator for HNSW indexes with separated vector storage.

This struct provides a unified interface for:

• Storing vectors in CollectionVectorStore
• Maintaining HNSW indexes via HnswIndexManager
• Coordinating insert/update/delete operations
• Rebuilding indexes from stored vectors

The coordinator ensures that:

1. Vectors are stored in the vector store for persistence
2. HNSW indexes are updated automatically for fast similarity search
3. Delete operations cascade properly to both stores

Implementations

impl<E: StorageEngine> VectorIndexCoordinator<E>

pub fn new(engine: E) -> Self

Create a new coordinator.

Arguments

• engine - The storage engine

pub fn with_manager(engine: E, index_manager: Arc<HnswIndexManager<E>>) -> Self

Create a coordinator with an existing index manager.

Use this when you need to share an index manager across multiple coordinators.

pub fn vector_store(&self) -> &CollectionVectorStore<E>

Get a reference to the vector store.

pub fn index_manager(&self) -> &Arc<HnswIndexManager<E>>

Get a reference to the index manager.

pub fn upsert_vector( &self, collection_id: CollectionId, entity_id: EntityId, collection_name: &str, vector_name: &str, data: &VectorData, ) -> Result<(), VectorError>

Upsert a vector, updating both storage and HNSW index.

This method:

1. Stores the vector in CollectionVectorStore
2. Updates the HNSW index (if one exists for this named vector)

Arguments

• collection_id - The collection ID
• entity_id - The entity ID
• collection_name - Collection name for index lookup
• vector_name - The vector name within the collection
• data - The vector data to store

Errors

Returns an error if storage or index update fails.

pub fn upsert_vectors_batch( &self, collection_id: CollectionId, collection_name: &str, vectors: &[(EntityId, &str, &VectorData)], ) -> Result<(), VectorError>

Upsert multiple vectors in a batch.

More efficient than calling upsert_vector multiple times as it batches storage operations and HNSW insertions.

Arguments

• collection_id - The collection ID
• collection_name - Collection name for index lookup
• vectors - List of (entity_id, vector_name, vector_data) tuples

pub fn delete_vector( &self, collection_id: CollectionId, entity_id: EntityId, collection_name: &str, vector_name: &str, ) -> Result<bool, VectorError>

Delete a vector from both storage and HNSW index.

Returns

Returns true if the vector was deleted from storage.

pub fn delete_entity_vectors( &self, collection_id: CollectionId, entity_id: EntityId, collection_name: &str, ) -> Result<usize, VectorError>

Delete all vectors for an entity.

This cascades the delete to all HNSW indexes for the collection.

Returns

Returns the number of vectors deleted from storage.

pub fn get_vector( &self, collection_id: CollectionId, entity_id: EntityId, vector_name: &str, ) -> Result<Option<VectorData>, VectorError>

Get a vector from storage.

pub fn get_all_vectors( &self, collection_id: CollectionId, entity_id: EntityId, ) -> Result<HashMap<String, VectorData>, VectorError>

Get all vectors for an entity.

pub fn search( &self, collection_name: &str, vector_name: &str, query: &Embedding, k: usize, ef_search: Option<usize>, ) -> Result<Vec<SearchResult>, VectorError>

Search for similar vectors using HNSW.

Arguments

• collection_name - The collection to search
• vector_name - The named vector to search
• query - The query embedding
• k - Number of results to return
• ef_search - Optional beam width (uses default if None)

Returns

A list of search results with entity IDs and distances.

pub fn search_with_filter<F>( &self, collection_name: &str, vector_name: &str, query: &Embedding, k: usize, predicate: F, ef_search: Option<usize>, ) -> Result<Vec<SearchResult>, VectorError>

where F: Fn(EntityId) -> bool,

Search with a filter predicate.

The predicate is applied during graph traversal, not as a post-filter.

pub fn rebuild_index_from_store( &self, collection_id: CollectionId, collection_name: &str, vector_name: &str, ) -> Result<usize, VectorError>

Rebuild an HNSW index from the vector store.

This is useful for:

• Recovering from index corruption
• Building an index for existing vectors
• Re-indexing after configuration changes

Arguments

• collection_id - The collection ID
• collection_name - Collection name for index lookup
• vector_name - The named vector to rebuild

Returns

The number of vectors indexed.

pub fn is_index_loaded(&self, collection: &str, vector_name: &str) -> bool

Check if an index is loaded in memory.

impl<E: StorageEngine> VectorIndexCoordinator<E>

Extension methods for the coordinator that require ownership of a storage engine.

pub fn create_index( &self, engine: E, collection: &str, vector_name: &str, dimension: usize, distance_metric: DistanceMetric, config: &HnswConfig, ) -> Result<String, VectorError>

Create an HNSW index for a collection’s named vector.

Note: This method takes a separate engine instance for index persistence. You may want to call this with a shared engine reference or a cloned engine.

Arguments

• engine - Storage engine for the index persistence
• collection - Collection name (e.g., “documents”)
• vector_name - Vector name within the collection (e.g., “text_embedding”)
• dimension - Vector dimension
• distance_metric - Distance metric for similarity
• config - HNSW configuration

Returns

The name of the created index (e.g., “documents_text_embedding_hnsw”)

pub fn drop_index( &self, engine: &E, index_name: &str, ) -> Result<bool, VectorError>

Drop an HNSW index.

pub fn drop_collection_indexes( &self, engine: &E, collection: &str, ) -> Result<Vec<String>, VectorError>

Drop all HNSW indexes for a collection.

pub fn has_index(&self, engine: &E, collection: &str, vector_name: &str) -> bool

Check if an HNSW index exists for a collection’s named vector.

pub fn load_index(&self, engine: E, index_name: &str) -> Result<(), VectorError>

Load an existing index into memory.

pub fn rebuild_index_from_scratch( &self, engine: E, collection_id: CollectionId, collection_name: &str, vector_name: &str, ) -> Result<usize, VectorError>

Rebuild an index from scratch with new configuration.

This drops the existing index data and creates a fresh index.

Arguments

• engine - Storage engine for the new index
• collection_id - The collection ID
• collection_name - Collection name
• vector_name - The named vector to rebuild

Returns

The number of vectors indexed.